def _core_(self, imgId):
img_info = self.coco.loadImgs(imgId)[0]
image_name = img_info['file_name']
print(image_name)
annIds = self.coco.getAnnIds(imgIds=img_info['id'],
catIds=self.catIds,
iscrowd=None)
anns = self.coco.loadAnns(annIds)
pseudomasks = []
height = img_info['height']
width = img_info['width']
pseudomasks = np.zeros((height, width), dtype=np.float64)
for ann in anns:
pointobb = ann['pointobb']
pseudomask = pointobb2pseudomask(height, width, pointobb)
pseudomasks += pseudomask
# self.progress_bar.update()
if save_np:
if self.add_seg:
return_flag = True
pseudomask_file = os.path.join(self.save_path, image_name)
stuff_things = cocoSegmentationToPng(self.coco,
imgId,
pseudomask_file,
vis=False,
return_flag=return_flag)
pseudomask_seg = (stuff_things[:, :, 0] + pseudomasks)
# pseudomask_seg_max = np.max(pseudomask_seg)
# pseudomask_seg_min = np.min(pseudomask_seg)
# pseudomask_seg = (pseudomask_seg - pseudomask_seg_min) / (pseudomask_seg_max - pseudomask_seg_min) * 255.0
pseudomask_seg = pseudomask_seg * 10.0
pseudomask_seg = pseudomask_seg.astype(np.uint8)
# pseudomasks_ = show_grayscale_as_heatmap(pseudomask_seg, True, return_img=True)
if return_flag:
# pseudomask_file = os.path.join(self.save_path, image_name.split('.png')[0])
# np.save(pseudomask_file, pseudomasks)
cv2.imwrite(pseudomask_file, pseudomask_seg)
else:
pseudomask_file = os.path.join(self.save_path,
image_name.split('.png')[0])
np.save(pseudomask_file, pseudomasks)
else:
image_file = os.path.join(self.imgDir, image_name)
img = cv2.imread(image_file)
pseudomask_file = os.path.join(self.save_path, image_name)
pseudomasks_ = show_grayscale_as_heatmap(pseudomasks,
False,
return_img=True)
alpha = 0.6
beta = (1.0 - alpha)
pseudomasks = cv2.addWeighted(pseudomasks_, alpha, img, beta, 0.0)
cv2.imwrite(pseudomask_file, pseudomasks)
return image_name
def __generate_pseudomask(self, imgId):
img_info = self.coco.loadImgs(imgId)[0]
image_name = img_info['file_name']
# if image_name != 'P2802__1.0__4914___4225.png':
# return
annIds = self.coco.getAnnIds(imgIds=img_info['id'], catIds=self.catIds, iscrowd=None)
anns = self.coco.loadAnns(annIds)
pseudomasks = []
height = img_info['height']
width = img_info['width']
area_map = np.ones((height, width)) * 1024 * 1024
pseudomasks = np.zeros((height, width), dtype=np.int32)
anchor_image = self.anchor_image[self.encode]
for ann in anns:
pointobb = ann['pointobb']
label_mask = self.coco.annToMask(ann) == 1
area = np.sum(label_mask)
transformed, mask_location = pointobb2pseudomask(pointobb, anchor_image, host_height = height, host_width = width)
transformed = transformed.astype(np.int32)
if self.method == 'min_area':
temp_pseudomasks = np.zeros((height, width), dtype=np.int32)
temp_pseudomasks[mask_location[1]:mask_location[3], mask_location[0]:mask_location[2]] = transformed
pseudomasks[label_mask] = np.where(area < area_map[label_mask], temp_pseudomasks[label_mask], pseudomasks[label_mask])
area_map[label_mask] = np.where(area < area_map[label_mask], area, area_map[label_mask])
elif self.method == 'min_score':
pseudomasks[mask_location[1]:mask_location[3], mask_location[0]:mask_location[2]] = np.where(transformed > pseudomasks[mask_location[1]:mask_location[3], mask_location[0]:mask_location[2]], transformed, pseudomasks[mask_location[1]:mask_location[3], mask_location[0]:mask_location[2]])
# save pseudomask
pseudomask_file = os.path.join(self.save_path, image_name)
pseudomasks = np.clip(pseudomasks, 0, 255)
# pseudomasks = pseudomasks * 255.0
pseudomasks = pseudomasks.astype(np.uint8)
cv2.imwrite(pseudomask_file, pseudomasks)
if self.save_vis:
image_file = os.path.join(self.imgDir, image_name)
img = cv2.imread(image_file)
pseudomask_vis_file = os.path.join(self.save_vis_path, image_name)
pseudomasks_ = show_grayscale_as_heatmap(pseudomasks / 255.0, self.show_pseudomask, return_img=True)
alpha = 0.6
beta = (1.0 - alpha)
pseudomasks = cv2.addWeighted(pseudomasks_, alpha, img, beta, 0.0)
cv2.imwrite(pseudomask_vis_file, pseudomasks)
264.6916809082031, 741.2076416015625, 82.70415496826172,
145.08529663085938, -0.42366891543622837
],
[
373.6033630371094, 713.7134399414062, 81.72265625,
179.64755249023438, -0.4245613407175139
]]
for thetaobb in thetaobbs:
pointobb = thetaobb2pointobb(thetaobb)
pointobbs.append(pointobb)
pseudomasks = np.zeros((height, width), dtype=np.int32)
for pointobb in pointobbs:
transformed, mask_location = pointobb2pseudomask(pointobb,
anchor_image,
host_height=height,
host_width=width)
# show_grayscale_as_heatmap(transformed)
transformed = transformed.astype(np.int32)
pseudomasks[mask_location[1]:mask_location[3],
mask_location[0]:mask_location[2]] = np.where(
transformed >
pseudomasks[mask_location[1]:mask_location[3],
mask_location[0]:mask_location[2]],
transformed,
pseudomasks[mask_location[1]:mask_location[3],
mask_location[0]:mask_location[2]])
pseudomasks_ = show_grayscale_as_heatmap(pseudomasks / 255.0,
False,
return_img=True)